Enterprise customer risk decision-making method and system based on deep learning

Through deep learning technology, customer feature information is collected and expanded, and combined with historical data, the problems of missing data and insufficient features in traditional risk decisions are solved, the accuracy and comprehensiveness of risk assessment are improved, and more flexible and accurate risk decisions are achieved.

CN119026918AInactive Publication Date: 2024-11-26SHENZHEN BORUI WORLD TECH CO LTD
View PDF 3 Cites 0 Cited by

Patent Information

Application Number
CN202411487917.9
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-10-24
Publication Date
2024-11-26
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

Traditional enterprise customer risk decisions are based on limited data and fixed models, resulting in missing customer data and insufficient characteristic information, which in turn reduces the accuracy of risk assessment and the one-sidedness of evaluation results, resulting in poor decision-making accuracy.

Method used

Using deep learning-based methods, customers' feature information is collected, missing information is screened, feature information is expanded, and risk prediction classification and analysis is carried out through deep learning, combined with historical maintenance data, comprehensive risk classification is calculated, and key features and missing types are identified, so as to Improve the accuracy and comprehensiveness of decision-making.

Benefits of technology

It improves the accuracy, comprehensiveness and flexibility of corporate customer risk decisions, enhances the ability to comprehensively evaluate customer risks, and reduces the one-sidedness of the evaluation results.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119026918A_ABST
    Figure CN119026918A_ABST
Patent Text Reader

Abstract

The invention discloses an enterprise customer risk decision-making method and system based on deep learning, and relates to the technical field of data processing, and the method comprises the steps: collecting a customer feature information set of a target customer to be subjected to risk analysis decision-making; screening to obtain a plurality of missing information types; obtaining a plurality of extended feature information sets; obtaining a plurality of first risk grades, and obtaining a plurality of second risk grades; obtaining a plurality of third risk grades, and obtaining a plurality of fourth risk grades; obtaining a comprehensive risk grade as a risk decision result; and analyzing and screening to obtain key feature information and a key missing type, and identifying a risk decision result. The technical problem that the accuracy of the enterprise customer risk decision is poorer due to the fact that the accuracy of risk assessment is reduced and the assessment result is one-sidedness caused by the lack of customer data and insufficient feature information in the existing enterprise customer risk decision is solved, and the technical effect of improving the accuracy, comprehensiveness and flexibility of the risk decision is achieved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the technical field related to data processing, and specifically to a method and system for enterprise customer risk decision-making based on deep learning. Background Art

[0002] In today's business environment, risk management of corporate customers is one of the key factors for the sustainable development and sound operation of enterprises. With the intensification of market competition and the rapid growth of the number of customers, enterprises are facing increasingly complex customer risk management challenges. However, traditional corporate customer risk decisions usually rely on a series of complex but relatively fixed risk assessment processes and tools to assess the credit risks of potential or existing customers. In the face of increasingly complex market environments and changing customer behaviors, its limitations and shortcomings are gradually highlighted. First, traditional corporate customer risk decisions are often based on limited data sets. When there is data loss, it is difficult to accurately assess risks, resulting in a failure to fully reflect the customer's true risk status. Secondly, the risk assessment model for traditional corporate customer risk decisions is based on historical data, which is relatively fixed and lacks flexibility and adaptability. When faced with new market environments and customer behaviors, it is difficult to accurately predict and assess risks.

[0003] Therefore, at the current stage, there are technical problems in the technologies related to enterprise customer risk decision-making, such as missing customer data and insufficient feature information, which in turn leads to reduced accuracy of risk assessment and one-sided assessment results, resulting in poor accuracy of enterprise customer risk decision-making. Summary of the invention

[0004] This application solves the technical problems of missing customer data and insufficient feature information in existing enterprise customer risk decision-making, which leads to reduced risk assessment accuracy and one-sided assessment results, resulting in poor accuracy of enterprise customer risk decision-making, by providing a deep learning-based enterprise customer risk decision-making method and system, thereby achieving the technical effect of improving the accuracy, comprehensiveness and flexibility of risk decision-making.

[0005] The present application provides a method for enterprise customer risk decision-making based on deep learning, the method comprising: collecting a customer feature information set of a target customer to be subject to risk analysis and decision-making in an enterprise database, the customer feature information set comprising feature information of multiple feature types; screening and obtaining multiple missing information types of missing feature information of the target customer based on the customer feature information set; performing feature information expansion processing on the customer feature information set to obtain multiple expanded feature information sets; performing risk prediction classification on the customer feature information set and the multiple expanded feature information sets based on deep learning to obtain multiple first risk classifications, and, based on the historical maintenance data of the target customer, classifying the customer feature information set and the multiple expanded feature information sets. The method comprises the following steps: performing risk analysis on the information set to obtain multiple second risk classifications; performing risk prediction classification on the multiple missing information types based on deep learning to obtain multiple third risk classifications, and performing risk prediction analysis on the multiple missing information types according to the historical maintenance data of the target customers to obtain multiple fourth risk classifications; calculating the multiple first risk classifications, multiple second risk classifications, multiple third risk classifications and multiple fourth risk classifications to obtain a comprehensive risk classification as a risk decision result; analyzing and screening to obtain key feature information and key missing types according to the multiple first risk classifications, multiple second risk classifications, multiple third risk classifications and multiple fourth risk classifications, and marking the risk decision result.

[0006] In a possible implementation, feature information expansion processing is performed on the customer feature information set to obtain multiple expanded feature information sets, and the following processing is performed: multiple feature information ranges of the multiple feature types are obtained; multiple feature information is expanded within the multiple feature information ranges according to a preset expansion ratio to obtain multiple expanded feature information intervals; in the enterprise database, multiple existing feature information sets that fall into the multiple expanded feature information intervals in the customer feature information sets of all customers are screened; and multiple existing feature information sets with the highest occurrence frequency are screened among the multiple existing feature information sets to obtain multiple expanded feature information sets.

[0007] In a possible implementation, risk prediction classification is performed on a customer feature information set and multiple extended feature information sets based on deep learning to obtain multiple first risk grades, and the following processing is performed: based on the enterprise database, a sample feature information set is collected, and a sample information risk grade set is collected according to the risk grade of the customer when different sample feature information appears; the sample feature information set and the sample information risk grade set are used to construct a risk predictor based on deep learning, wherein the risk predictor includes multiple risk prediction paths; the multiple feature information in the customer feature information set is input into the risk predictor, risk prediction classification is performed, multiple basic risk grade sets are obtained, and the basic risk grade with the highest frequency of occurrence is screened to obtain the first risk grade of multiple feature information; based on the deviation between the extended feature information in the multiple extended feature information sets and the corresponding feature information, multiple prediction ratio sets are determined, wherein the size of the deviation is negatively correlated with the size of the prediction ratio; based on the multiple prediction ratio sets, a corresponding proportion of risk prediction paths are randomly selected in the risk predictor, and the extended feature information in the multiple extended feature information sets is risk predicted and classified to obtain risk prediction results and the first risk grade of the multiple extended feature information sets is screened.

[0008] In a possible implementation, the sample feature information set and the sample information risk grading set are used to construct a risk predictor based on deep learning, and the following processing is also performed: a preset proportion of sample feature information and sample information risk grading are randomly extracted from the sample feature information set and the sample information risk grading set as a first prediction data set; the first prediction data set is used to construct a first risk prediction path based on deep learning and train it until convergence; the second prediction data set is continued to be randomly extracted, and the second risk prediction path is constructed and trained based on deep learning; and multiple risk prediction paths are continued to be constructed and integrated to obtain a risk predictor.

[0009] In a possible implementation, based on the historical maintenance data of the target customer, a risk analysis is performed on the customer feature information set and multiple extended feature information sets to obtain multiple second risk levels, and the following processing is also performed: based on the historical maintenance data of the target customer, the number of occurrences of the multiple feature information and the extended feature information in the multiple extended feature information sets, as well as the number of risk events that occur to the target customer when they occur, are obtained to obtain multiple feature information occurrence times and multiple information risk times; based on the multiple feature information occurrence times and multiple information risk times, multiple second risk levels are calculated.

[0010] In a possible implementation, risk prediction classification is performed on the multiple missing information types based on deep learning to obtain multiple third risk levels, and risk prediction analysis is performed on the multiple missing information types according to the historical maintenance data of the target customers to obtain multiple fourth risk levels. The following processing is also performed: based on the enterprise database, a set of sample missing information types is collected, and the risk levels of customers when different sample missing information types appear are collected to obtain a sample type risk level set; the sample missing information type set and the sample type risk level set are used to train a type risk predictor; the multiple missing information types are input into the type risk predictor for risk prediction analysis to obtain multiple third risk levels; according to the historical maintenance data of the target customers, the number of occurrences of the multiple missing information types and the number of risk events that occur to the customers when they appear are obtained to obtain multiple missing type occurrences and multiple type risk times, and multiple fourth risk levels are calculated.

[0011] In a possible implementation, based on the multiple first risk levels, multiple second risk levels, multiple third risk levels and multiple fourth risk levels, key characteristic information and key missing types are analyzed and screened, and the following processing is also performed: based on the multiple first risk levels and multiple second risk levels, multiple information risk levels are calculated; based on the multiple third risk levels and multiple fourth risk levels, multiple type risk levels are calculated; characteristic information with the largest information risk level and the missing information type with the largest type risk level are screened and obtained as key characteristic information and key missing types, and the risk decision result is marked.

[0012] This application also provides an enterprise customer risk decision-making system based on deep learning, including:

[0013] A customer feature information set collection module, which is used to collect customer feature information sets of target customers to be subject to risk analysis and decision making in the enterprise database, wherein the customer feature information set includes feature information of multiple feature types;

[0014] A missing information type screening module, wherein the multiple missing information type screening modules are used to screen and obtain multiple missing information types of the missing characteristic information of the target customer according to the customer characteristic information set;

[0015] An extended feature information set acquisition module, the extended feature information set acquisition module is used to perform feature information expansion processing on the customer feature information set to obtain multiple extended feature information sets;

[0016] The first and second risk classification acquisition modules are used to perform risk prediction classification on the customer feature information set and the multiple extended feature information sets based on deep learning to obtain multiple first risk classifications, and to perform risk analysis on the customer feature information set and the multiple extended feature information sets based on the historical maintenance data of the target customer to obtain multiple second risk classifications;

[0017] The third and fourth risk classification acquisition modules are used to perform risk prediction classification on the multiple missing information types based on deep learning to obtain multiple third risk classifications, and to perform risk prediction analysis on the multiple missing information types based on the historical maintenance data of the target customer to obtain multiple fourth risk classifications;

[0018] a comprehensive risk grading obtaining module, the comprehensive risk grading obtaining module being used to calculate the plurality of first risk gradings, the plurality of second risk gradings, the plurality of third risk gradings and the plurality of fourth risk gradings to obtain a comprehensive risk grading as a risk decision result;

[0019] A risk decision result identification module is used to analyze and screen the key feature information and key missing types according to the multiple first risk levels, multiple second risk levels, multiple third risk levels and multiple fourth risk levels, and identify the risk decision results.

[0020] The enterprise customer risk decision-making method and system based on deep learning proposed in this application collects customer feature information sets of target customers to be subject to risk analysis and decision-making in the enterprise database, including feature information of multiple feature types; screens and obtains multiple missing information types of missing feature information of target customers based on the customer feature information sets; performs feature information expansion processing on the customer feature information sets to obtain multiple expanded feature information sets; performs risk prediction and classification on the customer feature information sets and multiple expanded feature information sets based on deep learning to obtain multiple first risk classifications; and performs risk prediction and classification on the customer feature information sets and multiple expanded feature information sets based on the historical maintenance data of the target customers. Risk analysis is performed to obtain multiple second risk classifications; risk prediction classification is performed on multiple missing information types based on deep learning to obtain multiple third risk classifications; risk prediction analysis is performed on multiple missing information types based on historical maintenance data of target customers to obtain multiple fourth risk classifications; multiple first risk classifications, multiple second risk classifications, multiple third risk classifications and multiple fourth risk classifications are calculated to obtain comprehensive risk classifications as risk decision results; key feature information and key missing types are analyzed and screened based on multiple first risk classifications, multiple second risk classifications, multiple third risk classifications and multiple fourth risk classifications, and risk decision results are marked. The technical problem of missing customer data and insufficient feature information in existing enterprise customer risk decisions, which leads to reduced risk assessment accuracy and one-sided assessment results, making enterprise customer risk decisions less accurate, is solved, and the technical effect of improving the accuracy, comprehensiveness and flexibility of risk decisions is achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the technical solution of the embodiment of the present disclosure, the accompanying drawings of the embodiment of the present disclosure will be briefly introduced below. A flow chart is used in the present application to illustrate the operations performed by the system according to the embodiment of the present application. It should be understood that the preceding or following operations are not necessarily performed accurately in order. On the contrary, various steps can be processed in reverse order or simultaneously as needed. At the same time, other operations can also be added to these processes, or one or more operations can be removed from these processes.

[0022] Figure 1 A flowchart of a method for enterprise customer risk decision-making based on deep learning provided in an embodiment of the present application;

[0023] Figure 2 A schematic diagram of the structure of an enterprise customer risk decision-making system based on deep learning provided in an embodiment of the present application.

[0024] Explanation of the accompanying drawings: customer feature information set acquisition module 10, missing information type screening module 20, expanded feature information set acquisition module 30, first and second risk grading acquisition module 40, third and fourth risk grading acquisition module 50, comprehensive risk grading acquisition module 60, risk decision result identification module 70. DETAILED DESCRIPTION

[0025] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below.

[0026] In order to make the purpose, technical solutions and advantages of the present application clearer, the present application will be further described in detail below in conjunction with the accompanying drawings. The described embodiments should not be regarded as limiting the present application. All other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of this application.

[0027] In the following description, reference is made to "some embodiments", which describe a subset of all possible embodiments, but it is understood that "some embodiments" may be the same subset or different subsets of all possible embodiments, and may be combined with each other without conflict, and the terms "first\second" involved are merely to distinguish similar objects and do not represent a specific ordering of objects. The terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or modules that are not clearly listed or inherent to these processes, methods, products, or devices. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those generally understood by technicians in the technical field of this application. The terms used herein are for the purpose of describing the embodiments of the present application only.

[0028] The present application embodiment provides a method for enterprise customer risk decision-making based on deep learning, such as Figure 1 As shown, the method includes:

[0029] Step S100, in the enterprise database, collects a customer feature information set of the target customer to be subject to risk analysis decision, wherein the customer feature information set includes feature information of multiple feature types. The enterprise database is a database that stores, organizes and manages enterprise data according to a data structure, including various detailed information of many enterprises, such as registration information, financial information, employee information, production capacity, etc. The customer feature information set refers to a set of detailed information of the target customer that needs to be collected in the enterprise database in order to make a specific risk analysis decision, and the customer feature information set includes feature information of multiple feature types. Specifically, feature information of multiple feature types refers to multiple aspects of information that need to be considered and collected when constructing the target customer feature information set, such as basic information of enterprise customers, business information, transaction records and other categories.

[0030] Step S200, based on the customer characteristic information set, screen and obtain multiple missing information types of the characteristic information that is missing for the target customer. The multiple missing information types of the characteristic information that is missing for the target customer refer to information categories in which multiple data in the customer characteristic information set are missing or incomplete. The missing characteristic information of the target customer may be due to the fact that the customer has no public data, that is, in the customer characteristic information set, there are certain characteristic information fields or data items that are not filled in, cannot be obtained, or are empty, which is related to the risk of corporate customers and may include: missing financial information, such as the customer's bank statement, income certificate, asset certificate and other financial data missing; missing transaction records, such as the customer's historical purchase records, order details, payment information and other transaction data missing; missing basic information, such as the customer's name, address, contact information and other basic information is incomplete or missing; missing industry-specific data, professional qualification information, certification information, etc.

[0031] Step S300, feature information expansion processing is performed on the customer feature information set to obtain multiple expanded feature information sets. Feature information expansion processing is performed on the customer feature information set to obtain multiple expanded feature information sets, specifically, a series of technologies and methods are used to enhance, supplement and derive the original customer feature information set, so as to generate a richer and more predictive feature set, that is, to obtain multiple expanded feature information sets, for example, using mean, median, mode filling, or using machine learning algorithm prediction filling to handle missing values; using label encoding for categorical data, based on existing features, new features are calculated, and then unsupervised learning methods such as clustering and dimensionality reduction are used to discover hidden structures and patterns in the data, or supervised learning methods such as decision trees, random forests, neural networks, etc. are used to select features, and feature importance evaluation methods (such as model-based coefficients, tree-based model feature importance, etc.) are used to select the most predictive features, and finally multiple expanded feature information sets are obtained, which not only contain the information in the original data, but also expand new and more predictive features, which help to improve the accuracy and comprehensiveness of customer risk decision-making.

[0032] In a possible implementation, step S300 further includes step S310, obtaining multiple feature information ranges of the multiple feature types. For different types of features, determine the value range or description range of the information. It also includes step S320, according to a preset expansion ratio, within the multiple feature information ranges, the multiple feature information is extended to obtain multiple extended feature information intervals. The preset expansion ratio is a ratio set according to data analysis requirements before feature expansion, that is, within the value range of the feature information, the feature information is extended based on the preset expansion ratio. It also includes step S330, in the enterprise database, screening the customer feature information sets of all customers that fall into the multiple extended feature information intervals. The extended feature information interval is a new range or classification obtained by expanding the original feature information range according to a preset expansion ratio. Specifically, in the enterprise database, the customer feature information sets of all customers are traversed, and for each feature information in the feature information set of each customer, it is checked whether it falls into the defined multiple extended feature information intervals. If a certain feature information falls into a certain extended feature information interval, the entire customer feature information set of this customer is marked as an existing feature information set that meets the screening conditions, and finally multiple existing feature information sets are obtained.

[0033] Step S300 also includes step S340, in which multiple existing feature information sets with the highest frequency of occurrence are screened from the multiple existing feature information sets to obtain multiple extended feature information sets. For the multiple existing feature information sets that have been screened, the number of times or frequency of occurrence of each feature information set in the data set is counted by traversing the data set and calculating the count of each feature information set, and the frequency of occurrence of each feature information set obtained by counting is sorted, usually in descending order, and then, according to business needs, several feature information sets with the highest frequency of occurrence are selected, indicating that they appear frequently in the data set and may represent a certain customer group feature data or market trend, thereby obtaining multiple extended feature information sets.

[0034] Step S400, performing risk prediction classification on the customer feature information set and multiple extended feature information sets based on deep learning to obtain multiple first risk levels, and performing risk analysis on the customer feature information set and multiple extended feature information sets based on the historical maintenance data of the target customers to obtain multiple second risk levels. Based on deep learning technology, such as convolutional neural network (CNN), recurrent neural network (RNN), etc., a neural network model with multiple hidden layers is constructed to automatically learn and extract the characteristic information set of the target customer for analyzing and predicting the risk level of the customer. Specifically, the neural network model includes predictors of multiple different paths, and the predictors of all paths are used to learn the input customer characteristic information set and the expanded characteristic information set, and risk prediction classification is performed based on the deviation between the expanded characteristic information and the corresponding characteristic information, and the customer's risk level is output, that is, multiple first risk classifications; historical maintenance data refers to the target customer's past transaction records, behavioral data, historical risk level and other information in the system, which provides a historical perspective on the customer's risk status. Based on the historical maintenance data of the target customer, risk analysis is performed, which may include statistical analysis, time series analysis, machine learning and other methods to identify the customer's risk trend, periodic characteristics, abnormal behavior, etc. Through risk analysis, the target customer's risk level in a specific time period is obtained, that is, multiple second risk classifications, which are different from the first risk classification obtained based on the deep learning model and rely more on the customer's historical data and risk performance.

[0035] In a possible implementation, step S400 further includes step S410, collecting a sample feature information set based on the enterprise database, and collecting a sample information risk classification set according to the risk classification of the customer when different sample feature information appears. Feature data such as basic customer information, enterprise transaction history data, business conditions and patterns are screened out from the enterprise database to form a sample feature information set, and each element in the set represents a feature information set of an enterprise customer. Enterprises usually set their own risk classification standards based on the credit record, transaction behavior, historical default situation, etc. of the customer, such as high risk, low risk or unknown risk, etc., evaluate the feature information set of each customer, and determine its corresponding risk classification. Specifically, pair each sample feature information set with its corresponding risk classification to form a sample data point containing feature information and risk classification, and combine multiple such sample data points together to form a sample information risk classification set, and each element in the set is a pairing of a feature information set and a corresponding risk classification.

[0036] Step S400 also includes step S420, using the sample feature information set and the sample information risk classification set, based on deep learning, to construct a risk predictor, wherein the risk predictor includes multiple risk prediction paths. A prediction model is constructed based on an appropriate neural network architecture (such as a convolutional neural network, a recurrent neural network, or a deep belief network, etc.), and the sample feature information set and the sample information risk classification set are used as training data to train the prediction model to obtain a risk predictor, and the parameters of the model are continuously adjusted through optimization methods such as back propagation algorithm and gradient descent to minimize the error between the prediction result and the actual risk classification, wherein multiple risk prediction paths are set in the risk predictor, and the multiple risk prediction paths can be regarded as different branches inside the prediction model, which are used to process different types of feature information or perform different prediction tasks, and each risk prediction path includes a set of neural network layers, which are used to extract specific risk features from the input feature information and output corresponding risk prediction results. For example, for risk prediction containing multiple types of feature information, multiple parallel risk prediction paths are set, and each path corresponds to processing one type of feature information.

[0037] Step S400 also includes step S430, inputting the multiple characteristic information in the customer characteristic information set into the risk predictor, performing risk prediction classification, obtaining multiple basic risk classification sets, and screening the basic risk classification with the highest frequency, to obtain the first risk classification of multiple characteristic information. The multiple characteristic information in the customer characteristic information set is input into the risk predictor, and the risk predictor generates one or more basic risk classifications for each customer based on the input characteristic information through internal calculations, representing the risk level of the customer in different dimensions or different risk prediction paths, and performs statistics on the basic risk classifications of all customers, calculating the frequency of each basic risk classification in all prediction results, screening out the basic risk classification with the highest frequency from the statistical results, and using the screened basic risk classification with the highest frequency as the first risk classification of multiple characteristic information, and the first risk classification has high representativeness and accuracy.

[0038] Step S400 also includes step S440, determining multiple prediction ratio sets according to the deviations between the expanded feature information and the corresponding feature information in the multiple expanded feature information sets, wherein the size of the deviation is negatively correlated with the size of the prediction ratio. Ensure that each expanded feature information set has a corresponding original feature information set for the same customer sample, calculate the deviation between each feature in each expanded feature information set and the corresponding feature in the original feature information set, calculate the average or weighted average of all feature deviations for each expanded feature information set, which represents the overall deviation degree between the expanded feature information set and the original feature information set, and the smaller the deviation, the larger the prediction ratio should be. A smaller deviation means that the expanded feature information set is closer to the original feature information set, and can provide accurate prediction results. A mapping rule is established to map the deviation value to the prediction ratio value, and then a prediction ratio is determined using the mapping relationship according to the deviation value between each expanded feature information set and the original feature information set, and the prediction ratios of all expanded feature information sets are combined to form a prediction ratio set.

[0039] Step S400 also includes step S450, randomly selecting a corresponding proportion of risk prediction paths in the risk predictor according to the multiple prediction ratio sets, performing risk prediction classification on the extended feature information in the multiple extended feature information sets, obtaining risk prediction results and screening to obtain the first risk level of the multiple extended feature information sets. According to the prediction ratio set calculated previously, for each extended feature information set, a corresponding proportion of risk prediction paths are randomly selected in the risk predictor. For example, if the prediction ratio of a certain extended feature information set is 0.6, 60% of the paths are randomly selected from all available risk prediction paths, and the selected extended feature information set is input into the selected risk prediction path. Each path will calculate and process the input feature information according to its own structure and parameters to generate one or more risk prediction results. For each extended feature information set, the prediction results on all selected paths are weighted averaged to obtain a comprehensive prediction result, and the first risk level of each extended feature information set is determined.

[0040] In a possible implementation, step S420 further includes step S421, randomly extracting a preset proportion of sample feature information and sample information risk grading from the sample feature information set and sample information risk grading set as the first prediction data set. Randomly extract a preset proportion of sample feature information and sample information risk grading from the sample feature information set and sample information risk grading set. The preset proportion determines how many samples are extracted from the original data set as the first prediction data set, ensuring that the extracted sample feature information and its corresponding sample information risk grading maintain a one-to-one correspondence. When extracting samples, the randomness of the process must be ensured to ensure that the extracted samples are representative and can reflect the distribution of the original data set. The extracted sample feature information and the corresponding sample information risk grading are combined to form a first prediction data set. It also includes step S422, using the first prediction data set, based on deep learning, to construct a first risk prediction path and train to convergence. Determine the network structure of the prediction model, including the settings of the input layer, hidden layer and output layer. The input layer should match the dimension of the sample feature information in the first prediction data set. The output layer should be able to output the corresponding risk classification. The number of hidden layers and the number of neurons in each layer can be adjusted according to the complexity of the task and the scale of the data. The sample feature information in the first prediction data set is input into the model, and the output of the model (i.e., the predicted risk classification) is obtained by forward propagation calculation. The prediction result of the model is compared with the actual risk classification in the first prediction data set, and the loss function value is calculated. Then, the gradient of each parameter in the model is calculated by the back propagation algorithm, and then the optimization algorithm (such as gradient descent) is used to update the parameters of the model according to the gradient until the model reaches convergence or reaches a preset number of iterations, and finally the first risk prediction path is obtained. It also includes step S423, continuing to randomly extract the second prediction data set, and constructing and training the second risk prediction path based on deep learning. It also includes step S424, continuing to construct multiple risk prediction paths, and integrating to obtain a risk predictor.

[0041] In a possible implementation, step S400 further includes step S460, which obtains the occurrence times of the multiple feature information and the extended feature information in the multiple extended feature information sets, and the number of times the target customer has a risk event when the multiple feature information appears, according to the historical maintenance data of the target customer, to obtain the occurrence times of the multiple feature information and the multiple information risk times. The historical maintenance data of the target customer is traversed, the occurrence times of each feature information (including the original feature information and the extended feature information) are counted, and the number of risk events that occur to the target customer when these feature information appear is recorded, such as the number of overdue payments, complaint records, fraudulent behaviors, etc. Specifically, whenever a feature information appears in the historical record of the target customer, it is checked whether the record contains a risk event. If it does, the number of information risk events corresponding to the feature information is increased by 1, and finally a data set is obtained, which includes the occurrence times of each feature information (including the original feature information and the extended feature information) in the historical maintenance data of the target customer. Similarly, a data set is also obtained, which includes the number of times the target customer has a risk event when each feature information appears. The step S470 is also included, wherein multiple second risk levels are calculated based on the occurrence times of the multiple characteristic information and the multiple information risk times. The second risk classification is mainly based on the historical data and risk performance of the target customer.

[0042] Step S500, risk prediction classification is performed on the multiple missing information types based on deep learning to obtain multiple third risk classifications, and risk prediction analysis is performed on the multiple missing information types according to the historical maintenance data of the target customer to obtain multiple fourth risk classifications. In the customer feature data set, there are multiple types of missing feature information, which may affect the comprehensive risk assessment of the customer. For multiple missing information types, different situations of missing information (such as whether it is missing, the type of missing, the proportion of missing, etc.) can be input into the deep learning model as features. The learning of these features can predict the impact of missing information on the overall risk and give the corresponding risk level, i.e., the third risk classification; the historical maintenance data of the target customer is analyzed to identify the missing information types and the impact of these missing information types on the historical risk status of the customer. Based on the analysis results, risk prediction is performed on different missing information types to evaluate the risk level that they may cause in the future. According to the prediction results, multiple fourth risk classifications are given, which reflect the degree of risk that different missing information types may bring in the future. These two risk classifications provide risk assessment of missing information types from two different perspectives: model prediction and historical data, providing a more comprehensive and detailed basis for risk management.

[0043] In a possible implementation, step S500 further includes step S510, based on the enterprise database, collecting a set of sample missing information types, and collecting the risk grading of the customer when different sample missing information types appear, to obtain a sample type risk grading set. The enterprise database may contain a large amount of customer information, but in some cases, some information fields may be missing or unavailable, such as customer identity information, financial information, transaction records, credit history, etc. Specifically, all possible true information types are screened out from the enterprise database and organized into a set, namely, a sample missing information type set, and the risk grading corresponding to the customer when a specific missing information type appears in the database is analyzed and determined, and then a sample type risk grading set is obtained, each element in the set contains a missing information type and one or more risk gradings, indicating the risk level that the customer may be at when the missing information type appears. It also includes step S520, using the sample missing information type set and the sample type risk grading set to train a type risk predictor. The sample missing information type set and the sample type risk grading set are divided into training set, validation set and test set. The training set data is used to train the model. The model predicts the risk level corresponding to the new missing information type by learning the relationship between the input features (missing information type) and the target variable (risk level). The validation set is used to adjust the model parameters and hyperparameters, and the test set is used to evaluate the performance of the model, and finally a type risk predictor is obtained.

[0044] Step S500 also includes step S530, inputting the multiple missing information types into the type risk predictor for risk prediction analysis to obtain multiple third risk levels. The prepared missing information types are used as input data and input into the trained type risk predictor. The type risk predictor will perform risk prediction analysis based on the input missing information types and the knowledge and experience learned during its training process to evaluate the impact of different missing information types on the risk level. The type risk predictor will output a third risk level corresponding to each input missing information type, which represents the risk level that the customer may face based on the predicted missing information type. It also includes step S540, according to the historical maintenance data of the target customer, obtaining the number of occurrences of the multiple missing information types and the number of risk events that occurred when the customer appeared, obtaining multiple missing type occurrences and multiple type risk times, and calculating to obtain multiple fourth risk levels. The occurrence of all missing information types is collected from the historical maintenance data of the target customers. For each missing information type, the number of times it appears in the historical maintenance data is counted. At the same time, the number of risk events that occur to the target customers when the missing information type appears is recorded. For each missing information type, its risk probability is calculated. The risk probability is the ratio of the number of risk events that occur when the missing information type appears to the total number of occurrences of the missing information type. Then, according to the risk level classification standard, a risk level is assigned to each missing information type, namely the fourth risk level, which indicates the risk level corresponding to the missing information type.

[0045] Step S600, the multiple first risk classifications, multiple second risk classifications, multiple third risk classifications and multiple fourth risk classifications are calculated to obtain a comprehensive risk classification as a risk decision result. The risk classification results from different sources and based on different methods are integrated to obtain a more comprehensive and accurate comprehensive risk classification, which will be used as the final result of the risk decision to guide subsequent business decisions and risk management. For example, different weights are assigned to each risk classification according to its importance and reliability, and then they are weighted averaged to obtain a comprehensive score; or a machine learning model is trained to use each risk classification as an input feature to predict the comprehensive risk classification; the final comprehensive risk classification will be a more comprehensive and comprehensive evaluation result, which comprehensively considers the customer's current data status, historical performance, the impact of missing information and possible future risk trends.

[0046] Step S700: According to the multiple first risk classifications, multiple second risk classifications, multiple third risk classifications, and multiple fourth risk classifications, analyze and screen to obtain key feature information and key missing types, and mark the risk decision results. The risk classifications previously obtained by different methods are analyzed in depth, with the aim of identifying those key feature information and key missing types that have a significant impact on the risk decision results, and marking the final risk decision results accordingly. Specifically, a detailed analysis is performed on the multiple first risk classifications, the second risk classifications, the third risk classifications, and the fourth risk classifications, such as comparing the differences between different risk classifications, the changing trends, and the correlations between them and the final risk decision results. During the analysis process, feature information that has a greater impact on the risk decision results is identified. The key feature information may be certain specific attributes, behavior patterns, or transaction characteristics of the customer, which show obvious differences in different risk classifications. XOR trend; similarly, when analyzing risk grading, screen out missing information types and evaluate their impact on risk decision results. For example, some missing information types may have a significant impact on risk prediction due to their importance or missing frequency. By comparing the distribution of different missing types in different risk gradings, identify the key missing types that have a significant impact on risk decision results. After determining the key feature information and key missing types, mark the final risk decision results, for example, by adding additional labels, annotations or weights to the risk decision results. The content of the mark may include a description of the key feature information, an explanation of the key missing types, and their specific impact on the risk decision results.

[0047] In a possible implementation, step S700 further includes step S710, calculating and obtaining multiple information risk levels according to the multiple first risk levels and the multiple second risk levels. The multiple first risk levels and the multiple second risk levels are processed, for example, using the weighted average method, assigning different weights to each first risk level and the second risk level, and then calculating the weighted average as the information risk level, mapping the calculated information risk level results to the risk level classification standard of the enterprise customer, and determining the specific risk level corresponding to each information risk level, that is, obtaining multiple information risk levels. It also includes step S720, calculating and obtaining multiple type risk levels according to the multiple third risk levels and the multiple fourth risk levels. Similarly, the multiple third risk levels and the multiple fourth risk levels are calculated using the weighted average method, and the final result is multiple type risk levels, which integrates the dual information of predictive analysis and historical data. It also includes step S730, screening and obtaining the characteristic information with the largest information risk level and the missing information type with the largest type risk level as key characteristic information and key missing type, and marking the risk decision result. Among all the calculated information risk levels, the characteristic information with the highest level is screened out. These characteristic information are often the factors that have the greatest impact on the risk decision results. Similarly, among all the calculated type risk levels, the missing information types with the highest level are screened out. These missing information types often represent the factors that are most likely to cause risk events when missing. The maximum characteristic information and the maximum missing information type screened out are used as key characteristic information and key missing type to mark the risk decision results. Specifically, in the risk decision report, risk score or risk warning, these key characteristic information and key missing types are highlighted or emphasized, which helps to identify the source of risk more quickly and make more accurate risk decisions.

[0048] In the above, refer to Figure 1 The enterprise customer risk decision-making method based on deep learning according to an embodiment of the present invention is described in detail. Figure 2 The present invention describes an enterprise customer risk decision-making system based on deep learning according to an embodiment of the present invention.

[0049] The enterprise customer risk decision system based on deep learning according to the embodiment of the present invention is used to solve the technical problems of missing customer data and insufficient feature information in existing enterprise customer risk decisions, which in turn leads to reduced risk assessment accuracy and one-sided assessment results, making the accuracy of enterprise customer risk decisions poor, and achieves the technical effect of improving the accuracy, comprehensiveness and flexibility of risk decisions. The enterprise customer risk decision system based on deep learning includes: a customer feature information set acquisition module 10, a missing information type screening module 20, an expanded feature information set acquisition module 30, a first and second risk grading acquisition module 40, a third and fourth risk grading acquisition module 50, a comprehensive risk grading acquisition module 60, and a risk decision result identification module 70.

[0050] A customer feature information set collection module 10, which is used to collect customer feature information sets of target customers to be subject to risk analysis and decision making in an enterprise database, wherein the customer feature information set includes feature information of multiple feature types;

[0051] A missing information type screening module 20, wherein the multiple missing information type screening modules 20 are used to screen and obtain multiple missing information types of the missing characteristic information of the target customer according to the customer characteristic information set;

[0052] An extended feature information set acquisition module 30, the extended feature information set acquisition module 30 is used to perform feature information expansion processing on the customer feature information set to obtain multiple extended feature information sets;

[0053] The first and second risk classification acquisition modules 40 are used to perform risk prediction classification on the customer feature information set and the multiple extended feature information sets based on deep learning to obtain multiple first risk classifications, and to perform risk analysis on the customer feature information set and the multiple extended feature information sets based on the historical maintenance data of the target customer to obtain multiple second risk classifications;

[0054] A third and fourth risk classification obtaining module 50, the third and fourth risk classification obtaining module 50 is used to perform risk prediction classification on the multiple missing information types based on deep learning to obtain multiple third risk classifications, and to perform risk prediction analysis on the multiple missing information types according to the historical maintenance data of the target customer to obtain multiple fourth risk classifications;

[0055] A comprehensive risk grading obtaining module 60, the comprehensive risk grading obtaining module 60 is used to calculate the multiple first risk gradings, the multiple second risk gradings, the multiple third risk gradings and the multiple fourth risk gradings to obtain a comprehensive risk grading as a risk decision result;

[0056] The risk decision result identification module 70 is used to analyze and screen the key feature information and key missing types according to the multiple first risk levels, multiple second risk levels, multiple third risk levels and multiple fourth risk levels, and identify the risk decision results.

[0057] The specific configuration of the extended feature information set acquisition module 30 will be described in detail below. The extended feature information set acquisition module 30 further includes: obtaining multiple feature information ranges of the multiple feature types; performing extension processing on multiple feature information within the multiple feature information ranges according to a preset extension ratio to obtain multiple extended feature information intervals; screening multiple existing feature information sets that fall into the multiple extended feature information intervals in the customer feature information sets of all customers in the enterprise database; screening multiple existing feature information sets with the highest occurrence frequency in the multiple existing feature information sets to obtain multiple extended feature information sets.

[0058] The specific configuration of the first and second risk grading acquisition modules 40 will be described in detail below. The first and second risk grading acquisition modules 40 may further include: based on the enterprise database, collecting a sample feature information set, and collecting a sample information risk grading set according to the risk grading of the customer when different sample feature information appears; using the sample feature information set and the sample information risk grading set, based on deep learning, constructing a risk predictor, wherein the risk predictor includes multiple risk prediction paths; inputting multiple feature information in the customer feature information set into the risk predictor, performing risk prediction classification, obtaining multiple basic risk grading sets, and screening the basic risk grading with the highest frequency of occurrence to obtain the first risk grading of multiple feature information; determining multiple prediction ratio sets according to the deviation between the extended feature information and the corresponding feature information in the multiple extended feature information sets, wherein the size of the deviation is negatively correlated with the size of the prediction ratio; according to the multiple prediction ratio sets, randomly selecting a corresponding proportion of risk prediction paths in the risk predictor, performing risk prediction classification on the extended feature information in the multiple extended feature information sets, obtaining risk prediction results and screening to obtain the first risk level of multiple extended feature information sets.

[0059] The specific configuration of the first and second risk grading acquisition modules 40 will be described in detail below. The first and second risk grading acquisition modules 40 may further include: randomly extracting a preset proportion of sample feature information and sample information risk grading from the sample feature information set and the sample information risk grading set as a first prediction data set; using the first prediction data set, based on deep learning, constructing a first risk prediction path and training it to convergence; continuing to randomly extract a second prediction data set, based on deep learning, constructing and training a second risk prediction path; continuing to construct multiple risk prediction paths, and integrating to obtain a risk predictor.

[0060] The specific configuration of the first and second risk classification acquisition modules 40 will be described in detail below. The first and second risk classification acquisition modules 40 further include: obtaining the occurrence times of the multiple feature information and the extended feature information in the multiple extended feature information sets according to the historical maintenance data of the target customer, and the number of times the risk event occurs for the target customer when the event occurs, and obtaining the occurrence times of multiple feature information and multiple information risk times; and calculating and obtaining multiple second risk levels according to the occurrence times of the multiple feature information and the multiple information risk times.

[0061] The specific configuration of the third and fourth risk grading acquisition modules 50 will be described in detail below. The third and fourth risk grading acquisition modules 50 further include: based on the enterprise database, collecting a set of sample missing information types, and collecting the risk grading of customers when different sample missing information types appear, to obtain a sample type risk grading set; using the sample missing information type set and the sample type risk grading set to train a type risk predictor; inputting the multiple missing information types into the type risk predictor for risk prediction analysis to obtain multiple third risk gradings; according to the historical maintenance data of the target customer, obtaining the number of occurrences of the multiple missing information types and the number of risk events that occurred when the customer appeared, obtaining multiple missing type occurrences and multiple type risk times, and calculating to obtain multiple fourth risk levels.

[0062] The specific configuration of the risk decision result identification module 70 will be described in detail below. The risk decision result identification module 70 may further include: calculating and obtaining multiple information risk levels according to the multiple first risk levels and the multiple second risk levels; calculating and obtaining multiple type risk levels according to the multiple third risk levels and the multiple fourth risk levels; screening and obtaining the characteristic information with the largest information risk level and the missing information type with the largest type risk level as key characteristic information and key missing type, and identifying the risk decision result.

[0063] The deep learning-based enterprise customer risk decision system provided in the embodiment of the present invention can execute the deep learning-based enterprise customer risk decision method provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0064] Although the present application makes various references to certain modules in the system according to the embodiments of the present application, any number of different modules may be used and run on the user terminal and / or server, and the various units and modules included are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; in addition, the specific names of the functional units are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of the present invention.

[0065] The above specific implementations do not constitute a limitation on the protection scope of this application. It should be understood by those skilled in the art that various modifications, combinations and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions and improvements made within the spirit and principles of this application should be included in the protection scope of this application.

Claims

1. The enterprise customer risk decision-making method based on deep learning is characterized by: The method comprises: In the enterprise database, a customer feature information set of a target customer to be subject to risk analysis decision is collected, wherein the customer feature information set includes feature information of multiple feature types; According to the customer feature information set, multiple missing information types of the missing feature information of the target customer are obtained by screening; Performing feature information expansion processing on the customer feature information set to obtain multiple expanded feature information sets; Performing risk prediction classification on a customer feature information set and a plurality of extended feature information sets based on deep learning to obtain a plurality of first risk classifications, and performing risk analysis on the customer feature information set and the plurality of extended feature information sets according to historical maintenance data of the target customer to obtain a plurality of second risk classifications; performing risk prediction classification on the multiple missing information types based on deep learning to obtain multiple third risk classifications, and performing risk prediction analysis on the multiple missing information types based on historical maintenance data of the target customer to obtain multiple fourth risk classifications; Calculating the plurality of first risk levels, the plurality of second risk levels, the plurality of third risk levels, and the plurality of fourth risk levels to obtain a comprehensive risk level as a risk decision result; According to the multiple first risk classifications, multiple second risk classifications, multiple third risk classifications and multiple fourth risk classifications, key feature information and key missing types are obtained through analysis and screening, and the risk decision results are marked.

2. The enterprise customer risk decision-making method based on deep learning according to claim 1 is characterized in that: Performing feature information expansion processing on the customer feature information set to obtain multiple expanded feature information sets, including: Acquire multiple feature information ranges of the multiple feature types; According to a preset expansion ratio, within the range of the plurality of characteristic information, the plurality of characteristic information is expanded to obtain a plurality of extended characteristic information intervals; In the enterprise database, screening the customer feature information sets of all customers to find multiple existing feature information sets that fall into the multiple extended feature information intervals; A plurality of existing feature information sets with the highest occurrence frequency are screened from the plurality of existing feature information sets to obtain a plurality of expanded feature information sets.

3. The enterprise customer risk decision-making method based on deep learning according to claim 2 is characterized in that: The customer feature information set and multiple extended feature information sets are classified based on deep learning to predict risks and obtain multiple first risk classifications, including: Based on the enterprise database, a sample feature information set is collected, and according to the risk classification of the customer when different sample feature information appears, a sample information risk classification set is collected; Using the sample feature information set and the sample information risk classification set, based on deep learning, a risk predictor is constructed, wherein the risk predictor includes multiple risk prediction paths; Inputting multiple characteristic information in the customer characteristic information set into the risk predictor, performing risk prediction classification, obtaining multiple basic risk classification sets, and screening the basic risk classification with the highest frequency to obtain the first risk classification of the multiple characteristic information; Determining a plurality of prediction ratio sets according to deviations between the expanded feature information and the corresponding feature information in the plurality of expanded feature information sets, wherein the size of the deviation is negatively correlated with the size of the prediction ratio; According to the multiple prediction ratio sets, a corresponding ratio of risk prediction paths is randomly selected in the risk predictor, risk prediction classification is performed on the extended feature information in the multiple extended feature information sets, risk prediction results are obtained, and the first risk level of the multiple extended feature information sets is screened.

4. The enterprise customer risk decision-making method based on deep learning according to claim 3 is characterized in that: The sample feature information set and the sample information risk classification set are used to construct a risk predictor based on deep learning, including: Randomly extracting a preset proportion of sample feature information and sample information risk classification from the sample feature information set and sample information risk classification set as a first prediction data set; Using the first prediction data set, based on deep learning, constructing a first risk prediction path and training it until convergence; Continue to randomly extract the second prediction data set, and build and train the second risk prediction pathway based on deep learning; Continue to construct multiple risk prediction pathways and integrate them to obtain a risk predictor.

5. The enterprise customer risk decision-making method based on deep learning according to claim 3 is characterized in that: According to the historical maintenance data of the target customer, risk analysis is performed on the customer feature information set and multiple extended feature information sets to obtain multiple second risk classifications, including: According to the historical maintenance data of the target customer, the occurrence times of the multiple feature information and the extended feature information in the multiple extended feature information sets, and the number of risk events of the target customer when they occur are obtained, and the occurrence times of multiple feature information and multiple information risk times are obtained; A plurality of second risk levels are calculated based on the plurality of feature information occurrence times and the plurality of information risk times.

6. The enterprise customer risk decision-making method based on deep learning according to claim 1 is characterized in that: Performing risk prediction classification on the multiple missing information types based on deep learning to obtain multiple third risk classifications, and performing risk prediction analysis on the multiple missing information types according to the historical maintenance data of the target customer to obtain multiple fourth risk classifications, including: Based on the enterprise database, a set of sample missing information types is collected, and risk grading of customers when different types of sample missing information appear is collected to obtain a set of sample type risk grading; Using the sample missing information type set and the sample type risk classification set, training a type risk predictor; Inputting the multiple missing information types into the type risk predictor to perform risk prediction analysis to obtain multiple third risk classifications; According to the historical maintenance data of the target customer, the number of occurrences of the multiple missing information types and the number of risk events that occurred when the customers occurred are obtained, the number of occurrences of multiple missing types and the number of risk of multiple types are obtained, and multiple fourth risk levels are calculated.

7. The enterprise customer risk decision-making method based on deep learning according to claim 1 is characterized in that: According to the plurality of first risk classifications, the plurality of second risk classifications, the plurality of third risk classifications and the plurality of fourth risk classifications, analyzing and screening to obtain key feature information and key missing types, including: Calculating and obtaining multiple information risk levels according to the multiple first risk levels and the multiple second risk levels; Calculating and obtaining multiple types of risk levels according to the multiple third risk levels and the multiple fourth risk levels; The characteristic information with the largest information risk classification and the missing information type with the largest type risk level are screened and used as key characteristic information and key missing type to mark the risk decision result.

8. The enterprise customer risk decision-making system based on deep learning is characterized by: The system is used to implement the enterprise customer risk decision-making method based on deep learning according to any one of claims 1 to 7, and the system includes: A customer feature information set collection module, which is used to collect customer feature information sets of target customers to be subject to risk analysis and decision making in the enterprise database, wherein the customer feature information set includes feature information of multiple feature types; A missing information type screening module, wherein the multiple missing information type screening modules are used to screen and obtain multiple missing information types of the missing characteristic information of the target customer according to the customer characteristic information set; An extended feature information set acquisition module, the extended feature information set acquisition module is used to perform feature information expansion processing on the customer feature information set to obtain multiple extended feature information sets; The first and second risk classification acquisition modules are used to perform risk prediction classification on the customer feature information set and the multiple extended feature information sets based on deep learning to obtain multiple first risk classifications, and to perform risk analysis on the customer feature information set and the multiple extended feature information sets based on the historical maintenance data of the target customer to obtain multiple second risk classifications; The third and fourth risk classification acquisition modules are used to perform risk prediction classification on the multiple missing information types based on deep learning to obtain multiple third risk classifications, and to perform risk prediction analysis on the multiple missing information types based on the historical maintenance data of the target customer to obtain multiple fourth risk classifications; a comprehensive risk grading obtaining module, the comprehensive risk grading obtaining module being used to calculate the plurality of first risk gradings, the plurality of second risk gradings, the plurality of third risk gradings and the plurality of fourth risk gradings to obtain a comprehensive risk grading as a risk decision result; A risk decision result identification module is used to analyze and screen the key feature information and key missing types according to the multiple first risk levels, multiple second risk levels, multiple third risk levels and multiple fourth risk levels, and identify the risk decision results.

Citation Information

Patent Citations

  • Two-stage credit risk assessment method and system suitable for fragmented data

    CN116029821A

  • Risk identification method and device and electronic equipment

    CN117764706A

  • Algorithm and models for creditworthiness based on user entered data within financial management application

    US11037236B1